Method, device, medium and program product for generating graph embedding vectors

Abstract

The present disclosure provides a method, device, medium and program product for generating graph embedding vectors, relating to the technical fields of artificial intelligence such as deep learning, computer vision and natural language processing. An embodiment of the method comprises: obtaining behavior information of a target user with respect to information, wherein the behavior information comprises target query information and behavior type information; constructing an information-related graph according to the behavior information, wherein the nodes of the information-related graph comprise a target user node and a target query information node, and the weights of the edges of the information-related graph are determined based on the behavior type information; performing graph embedding on the information-related graph to obtain feature vectors corresponding to the target user node and the target query information node in the information-related graph.

Description

Technical Field

[0001] This disclosure relates to the field of computer science, specifically to deep learning, computer vision, and natural language processing, and particularly to a method, apparatus, medium, and program product for generating graph embedding vectors. Background Technology

[0002] With the continuous development of deep learning, deep learning models have been widely applied to recommendation algorithms. In the current basic framework of deep learning recommendation systems, embedding methods mainly serve as embedding layers in deep learning networks, transforming high-dimensional sparse feature vectors into low-dimensional dense feature vectors. Among these, deep learning methods for graph embedding belong to graph neural networks. Because graph structures are irregular, each graph consists of multiple nodes and edges connecting them, and the number of neighboring nodes for each node is constantly changing, it is difficult to collect a fixed order and number of neighboring nodes when extracting graph embedding features, resulting in inaccurate extracted graph embedding features. Summary of the Invention

[0003] This disclosure provides a method, apparatus, medium, and program product for generating graph embedding vectors.

[0004] In a first aspect, embodiments of this disclosure propose a method for generating graph embedding vectors, comprising: obtaining behavioral information of a target user regarding information, wherein the behavioral information includes target query information and behavioral type information; constructing an information-related graph based on the behavioral information, wherein the nodes of the information-related graph include target user nodes and target query information nodes, and the weights of the edges of the information-related graph are determined based on the behavioral type information; and performing graph embedding on the information-related graph to obtain feature vectors corresponding to the target user nodes and target query information nodes in the information-related graph.

[0005] Secondly, embodiments of this disclosure propose a graph embedding vector generation apparatus, comprising: an information acquisition module configured to acquire behavioral information of a target user regarding information, wherein the behavioral information includes target query information and behavioral type information; a graph construction module configured to construct an information-related graph based on the behavioral information, wherein the nodes of the information-related graph include target user nodes and target query information nodes, and the weights of the edges of the information-related graph are determined based on the behavioral type information; and a graph embedding module configured to perform graph embedding on the information-related graph to obtain feature vectors corresponding to the target user nodes and target query information nodes in the information-related graph.

[0006] Thirdly, embodiments of this disclosure provide an electronic device, including: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the method as described in the first aspect.

[0007] Fourthly, embodiments of this disclosure provide a non-transitory computer-readable storage medium storing computer instructions for causing a computer to perform the methods described in the first aspect.

[0008] Fifthly, embodiments of this disclosure provide a computer program product including a computer program that, when executed by a processor, implements the method described in the first aspect.

[0009] The graph embedding vector generation method, apparatus, medium, and program product provided in this disclosure construct an information-related graph based on the target user's behavioral information regarding information, including target query information and behavioral type information. The nodes of this information-related graph include target user nodes and target query information nodes, and the edge weights are determined based on the behavioral type information, which enables the identification of target user nodes and target query information nodes. Then, graph embedding is performed on the information-related graph, which improves the accuracy of obtaining feature vectors.

[0010] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this disclosure, nor is it intended to limit the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description

[0011] Other features, objects, and advantages of this disclosure will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings. The drawings are provided for a better understanding of the invention and are not intended to limit the scope of this disclosure. Wherein:

[0012] Figure 1 This is an exemplary system architecture diagram to which this disclosure can be applied;

[0013] Figure 2 This is a flowchart of an embodiment of the graph embedding vector generation method according to the present disclosure;

[0014] Figure 3 This is a flowchart of an embodiment of the graph embedding vector generation method according to the present disclosure;

[0015] Figure 4 This is a flowchart of an embodiment of the graph embedding vector generation method according to the present disclosure;

[0016] Figure 5 This illustrates an application scenario where targeted information is delivered to a first user.

[0017] Figure 6 This is an application scenario of the graph embedding vector generation method disclosed herein;

[0018] Figure 7 This is a schematic diagram of the structure of an embodiment of the graph embedding vector generation apparatus according to the present disclosure;

[0019] Figure 8 This is a block diagram of an electronic device used to implement embodiments of the present disclosure. Detailed Implementation

[0020] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0021] It should be noted that, unless otherwise specified, the embodiments and features described in this disclosure can be combined with each other. This disclosure will now be described in detail with reference to the accompanying drawings and embodiments.

[0022] Figure 1 An exemplary system architecture 100 is shown, in which embodiments of the graph embedding vector generation method and apparatus of this disclosure can be applied.

[0023] like Figure 1 As shown, system architecture 100 may include terminal devices 101, 102, and 103, a network 104, and a server 105. Network 104 serves as the medium for providing communication links between terminal devices 101, 102, and 103 and server 105. Network 104 may include various connection types, such as wired or wireless communication links, or fiber optic cables, etc.

[0024] Various client applications and intelligent interactive applications, such as information processing applications and information browsing applications, can be installed on terminal devices 101, 102, and 103.

[0025] Terminal devices 101, 102, and 103 can be either hardware or software. When terminal devices 101, 102, and 103 are hardware, they can be electronic products that allow human-computer interaction with users through one or more methods such as a keyboard, touchpad, touchscreen, remote control, voice interaction, or handwriting devices. Examples include PCs (Personal Computers), mobile phones, smartphones, PDAs (Personal Digital Assistants), wearable devices, PPCs (Pocket PCs), tablets, smart car systems, smart TVs, smart speakers, laptops, and desktop computers, etc. When terminal devices 101, 102, and 103 are software, they can be installed in the aforementioned electronic devices. They can be implemented as multiple software programs or software modules, or as a single software program or software module. No specific limitations are made here.

[0026] Server 105 can obtain the target user's behavioral information regarding information, including target query information and behavioral type information; based on the behavioral information, it constructs an information relevance graph, wherein the nodes of the information relevance graph include target user nodes and target query information nodes, and the weights of the edges of the information relevance graph are determined based on the behavioral type information; it performs graph embedding on the information relevance graph to obtain the feature vectors corresponding to the target user nodes and target query information nodes in the information relevance graph.

[0027] Server 105 can be either hardware or software. When server 105 is hardware, it can be implemented as a distributed server cluster consisting of multiple servers, or as a single server. When server 105 is software, it can be implemented as multiple software programs or software modules (e.g., used to provide distributed services), or as a single software program or software module. No specific limitations are made here.

[0028] It should be noted that the graph embedding vector generation method provided in this embodiment is generally executed by server 105, and correspondingly, the graph embedding vector generation device is generally located in server 105.

[0029] It should be understood that Figure 1 The number of electronic devices, networks, and servers shown is merely illustrative. Any number of electronic devices, networks, and servers can be included depending on implementation needs.

[0030] Continue to refer to Figure 2 The diagram illustrates a flow 200 of an embodiment of a graph embedding vector generation method according to the present disclosure. The graph embedding vector generation method may include the following steps:

[0031] Step 201: Obtain the target user's behavioral information regarding the information, whereby the behavioral information includes the target query information and the behavioral type information.

[0032] In this embodiment, the execution body of the graph embedding vector generation method (e.g.) Figure 1 The terminal devices 101, 102, and 103 shown can obtain user behavior information related to information locally, or the execution entity of the graph embedding vector generation method (e.g., Figure 1 The server 105 shown is located locally or via a network (e.g., Figure 1 The network 104 shown above starts from the terminal device (e.g., Figure 1 The terminal devices 101, 102, and 103 shown acquire user behavior information related to the information. This behavior information can be the information corresponding to the actions taken by the target user on the retrieved information.

[0033] Here, the aforementioned behavioral information may include target query information and behavior type information. The target query information may be the information entered by the target user to query information; the behavior type information may be the information corresponding to the operation performed on the queried information.

[0034] The collection, storage, use, processing, transmission, provision, and disclosure of behavioral information involved in the technical solution disclosed herein all comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0035] Step 202: Construct an information relevance graph based on the behavioral information. The nodes of the information relevance graph include target user nodes and target query information nodes. The weights of the edges of the information relevance graph are determined based on the behavioral type information.

[0036] In this embodiment, the aforementioned executing entity can construct an information correlation graph based on behavioral information. This information correlation graph can include nodes and edges. Nodes can include target user nodes and target query information nodes, and the weights of the edges can be determined based on behavioral type information. The behavioral type information can be information corresponding to the type of operation performed by the target user on the information, such as the timing of the operation, the frequency of the operation, and the type of the operation. The type of operation can include at least one of the following: clicking, browsing, forwarding, etc.

[0037] Step 203: Perform graph embedding on the information correlation graph to obtain the feature vectors corresponding to the target user node and the target query information node in the information correlation graph.

[0038] In this embodiment, the aforementioned execution entity can perform graph embedding on the information-related graph to obtain the feature vectors corresponding to the target user node and the target query information node in the information-related graph. The above-mentioned graph embedding is also known as network embedding or network representation learning. Its purpose is to project each node in the information-related graph into a low-dimensional vector space, that is, to learn an effective representation or encoding for the information-related graph. These representations or encodings are the "embedding" of the information-related graph.

[0039] Here, the information-related graph can be embedded using Word2Vec (word to vector) to obtain the feature vectors corresponding to each node in the information-related graph. Furthermore, graph embedding also includes Node2Vec (node ​​to vector), and those skilled in the art can adopt graph embedding methods suitable for the scenario described in this disclosure, according to actual needs.

[0040] The graph embedding vector generation method provided in this embodiment constructs an information-related graph based on the target user's behavioral information regarding information, including target query information and behavioral type information. The nodes of the information-related graph include target user nodes and target query information nodes, and the edge weights are determined based on the behavioral type information, which can identify the target user nodes and target query information nodes. Then, graph embedding is performed on the information-related graph, which improves the accuracy of obtaining feature vectors.

[0041] Further reference Figure 3 , Figure 3 A flow 300 of an embodiment of a graph embedding vector generation method according to the present disclosure is shown. The graph embedding vector generation method may include the following steps:

[0042] Step 301: Obtain the target user's behavioral information regarding the information, wherein the behavioral information includes target query information and behavior type information.

[0043] In this embodiment, the execution body of the graph embedding vector generation method (e.g.) Figure 1 The terminal devices 101, 102, and 103 shown can obtain user behavior information related to information locally, or the execution entity of the graph embedding vector generation method (e.g., Figure 1 The server 105 shown is located locally or via a network (e.g., Figure 1 The network 104 shown above starts from the terminal device (e.g., Figure 1 The terminal devices 101, 102, and 103 shown acquire user behavior information related to the information. This behavior information can be the information corresponding to the actions taken by the target user on the retrieved information.

[0044] Step 302: Construct an information relevance graph based on the behavioral information. The nodes of the information relevance graph include target user nodes and target query information nodes. The weights of the edges of the information relevance graph are determined based on the behavioral type information.

[0045] In this embodiment, the aforementioned executing entity can construct an information correlation graph based on behavioral information. This information correlation graph can include nodes and edges. Nodes can include target user nodes and target query information nodes, and the weights of the edges can be determined based on behavioral type information. The behavioral type information can be information corresponding to the type of operation performed by the target user on the information, such as the timing of the operation, the frequency of the operation, and the type of the operation. The type of operation can include at least one of the following: clicking, browsing, forwarding, etc.

[0046] Step 303: Perform a random walk on the information correlation graph to obtain the target user's behavior sequence for the target query information.

[0047] In this embodiment, the execution entity can take the target user node in the information correlation graph as the starting point and the target query information node as the ending node, and walk through the information correlation graph to obtain the target user's behavior sequence for the target query information.

[0048] Step 304: Perform vector embedding on the behavior sequence to obtain the feature vectors corresponding to the target user node and the target query information node in the information correlation graph.

[0049] In this embodiment, the aforementioned execution entity can perform vector embedding on the behavior sequence to obtain the feature vectors corresponding to the target user node and the target query information node in the information correlation graph.

[0050] Here, the aforementioned vector embedding can be used to convert behavior sequences into feature vectors. The vector embedding described above can convert behavior sequences into feature vectors using word2vec (word to vector). Furthermore, vector embedding also includes Node2Vec (node ​​to vector), and those skilled in the art can adopt vector embedding methods suitable for the scenario described herein, according to actual needs.

[0051] In this embodiment, the specific operations of steps 301 and 302 have been described. Figure 2 Steps 201 and 202 in the illustrated embodiments are described in detail and will not be repeated here.

[0052] from Figure 3 It can be seen from this that, with Figure 2Compared to the corresponding embodiments, the graph embedding vector generation method in this embodiment emphasizes the step of graph embedding of the information-related graph to obtain the feature vectors corresponding to the target user nodes and target query information nodes in the information-related graph. Therefore, the scheme described in this embodiment can perform a random walk on the information-related graph to obtain the target user's behavior sequence for the target query information; then, it performs vector embedding on the behavior sequence to obtain the feature vectors corresponding to the target user nodes and target query information nodes in the information-related graph. Based on the target user's behavior information regarding the information, including the target query information and behavior type information, an information-related graph is constructed. The nodes of this information-related graph include target user nodes and target query information nodes, and the edge weights are determined based on the behavior type information, which can identify the target user nodes and target query information nodes; then, graph embedding is performed on the information-related graph, improving the accuracy of obtaining the feature vectors.

[0053] In some optional implementations of this embodiment, performing a random walk on the information relevance graph to obtain the target user's behavior sequence for the target query information includes: dividing the information relevance graph into multiple target information relevance graphs; and for each target information relevance graph, performing a random walk on each target information relevance graph to obtain the target user's behavior sequence for the target query information.

[0054] In this implementation, the aforementioned execution entity can divide the large information correlation graph into multiple smaller information correlation graphs (i.e., target information correlation graphs); then, a random walk is performed on each target information correlation graph to obtain the target user's behavioral sequence for the target query information.

[0055] In one example, the aforementioned execution entity can use the target user node as the starting node in an information correlation graph and the target query information node as the ending node. Since the entire information correlation graph is relatively large, considering performance consumption issues, the large information correlation graph can be divided into multiple smaller information correlation graphs. A random walk algorithm is used to walk through each smaller information correlation graph to obtain the behavior sequence of the target user node towards the target query information node. Then, we embed vectors into each sampled behavior sequence to obtain the feature vectors corresponding to the target user node and the target query information node in the information correlation graph. Through this information correlation graph, the interest set among various target user nodes can be captured.

[0056] It should be noted that the large information-related graph can be divided into a corresponding number of smaller information-related graphs based on the accuracy of subsequent information pushes; or the large information-related graph can be divided into a corresponding number of smaller information-related graphs based on work experience or randomly.

[0057] In this implementation, the large information correlation graph can be divided into multiple smaller information correlation graphs. Then, a random walk is performed on each target information correlation graph to obtain the target user's behavior sequence for the target query information, thereby reducing the performance overhead caused by performing random walks on the large information correlation graph.

[0058] In some optional implementations of this embodiment, the behavior type information includes at least one of the following: the timing of the operation, the frequency of the operation, and the type of the operation.

[0059] In this implementation, the timing of the above operations can be the order in which the target user interacts with the information. The frequency of the above operations can be the number of times the target user interacts with the information. The type of the above operations can be the way the user interacts with the information, such as clicking, browsing, forwarding, etc.

[0060] In this implementation, the weights of the edges in the information-related graph can be determined based on at least one of the following: the timing of the operation, the frequency of the operation, and the type of the operation, which are included in the behavior type information.

[0061] In some optional implementations of this embodiment, if the behavior type information includes the timing of the operation, the frequency of the operation, and the type of the operation, the weights of the edges of the information-related graph can be determined based on the following steps: determining the weights of the edges of the information-related graph based on the weights corresponding to the timing of the operation, the frequency of the operation, and the type of the operation.

[0062] In one example, the weight S of the edges in the information-related graph is:

[0063] s = a1e -x +a2e -y

[0064] Where s is the weight of each edge in the information-related graph, a is the weight of time series and frequency, x is the time series value, and y is the frequency value.

[0065] In this implementation, considering the differences in how different target user nodes express the same thing, the interests between various target user nodes can be effectively captured by setting the weights of the aforementioned behavior type information.

[0066] In some optional implementations of this embodiment, before constructing the information correlation graph, the method for generating the graph embedding vector may further include: classifying the target query information according to the business type to which the target query information belongs; and constructing the information correlation graph based on the behavioral information, which may include: constructing the information correlation graph based on the classified query information and behavioral type information.

[0067] In this implementation, the aforementioned executing entity can classify the target query information according to the business type to which the target query information belongs; then, an information correlation graph can be constructed based on the classified query information and behavior type information. The aforementioned business type can be the type to which the entity query information in the target query information belongs, such as electronic product type, office product type, etc. Entity terms refer to things or concepts with specific meaning in the target query information; for example, entity terms can include proper nouns such as names of people, places, organizations, and products.

[0068] It should be noted that the above entity query information can be the entity query information corresponding to the entity words in the target query information.

[0069] In this implementation, the target query information can be classified based on the business type to which it belongs, and then an information correlation graph can be constructed based on the classified query information and behavior type information.

[0070] In some optional implementations of this embodiment, classifying the target query information according to the business type to which the target query information belongs may include: determining the business type to which the target query information belongs based on a pre-trained classification model; and classifying the target query information according to the business type to which the target query information belongs.

[0071] In this implementation, the aforementioned execution entity can determine the business type to which the target query information belongs based on a pre-trained classification model; and then classify the target query information according to the business type.

[0072] In this implementation, the aforementioned execution entity can classify the business type of the target query information according to the pre-trained classification model, so as to select the target query information of suitable business types and construct an information correlation graph.

[0073] In some optional implementations of this embodiment, the pre-trained classification model can be determined based on the following steps: in response to the target query information's character length meeting a preset length threshold, entity requirement constraint information is obtained; the entity requirement constraint information and the target query information are used as training samples to optimize and train the pre-trained classification model.

[0074] In this implementation, the execution entity can determine whether to supplement the target query information with entity requirement limiting information based on the character length of the target query information. This allows for a more accurate representation of the meaning of the target query information, thereby optimizing the pre-trained classification model. The character length can be the number of characters in the target query information. The entity requirement limiting information can be used to supplement and limit the entity query information in the target query information, making the meaning presented to the execution entity more accurate. For example, the entity query information is "Product A," and the "entity requirement limiting information" is "Generated by Manufacturer B."

[0075] Here, the above-mentioned character length meeting the preset length threshold can be a character length less than or equal to the preset length threshold. This preset length threshold can be based on the classification accuracy of the classification model or set by the user based on their work experience.

[0076] In one example, new training samples can also be obtained to optimize a pre-trained classification model.

[0077] It should be noted that before optimizing the pre-trained classification model, it can also be trained using the target query information as samples and the corresponding business type as labels to obtain a classification model. The aforementioned classification models can be used to determine the business type to which the target query information belongs; for example, BERT (Bidirectional Encoder Representation from Transformers) and Distill-BERT.

[0078] In this implementation, the execution entity can obtain entity requirement limitation information when the character length of the target query information meets a preset length threshold; then, the entity requirement limitation information and the target query information are used as training samples to optimize and train the pre-trained classification model.

[0079] In some optional implementations of this embodiment, the target query information includes core keywords.

[0080] In this implementation, the target query information may include core terms, such as entity terms in the target query information.

[0081] It should be noted that the above solutions involving target query information and / or first query information can be replaced with the core keywords corresponding to that query information, which will not be elaborated further here.

[0082] In one example, taking the target query information as an example, the core words in the target query information can be determined based on the following steps:

[0083] The first method involves identifying each n-gram (Chinese language model) in the target query information and determining the weight of each n-gram based on its occurrence in the target query information, thereby obtaining the core words in the target query information.

[0084] When determining the weight of each n-gram, the following two methods may be used, including but not limited to: Method 1: Assign weights to each n-gram based on its term frequency (TF) in the target query information and the corresponding n value. Method 2: Assign weights to n-grams based on the number of sentences in the target query information and the inverse document frequency (IDF) of the n-gram.

[0085] The second method involves segmenting the target query information into words and removing stop words. Then, the word frequency of each word after stop word removal is calculated. Words with frequencies higher than a preset frequency threshold are identified, and weights are assigned to each identified word based on its frequency to obtain the core words. The higher the word frequency, the greater the weight of that word.

[0086] The third method involves segmenting the target query information into words and removing stop words. Then, the TF and IDF of each word after stop word removal are calculated. Words with TF-IDF values ​​higher than a preset TF-IDF threshold are identified, and weights are assigned to these words based on their TF-IDF values ​​to obtain the core word vectors. The larger the TF-IDF value, the greater the weight of the word.

[0087] The fourth method is to segment the target query information and remove stop words, then assign weights to each word based on the number of sentences in the target query information where each word appears after removing stop words and the IDF of each word. Select words with weight values ​​higher than the preset weight threshold to obtain the core word vector.

[0088] Further reference Figure 4 , Figure 4 A flow 400 is shown as an embodiment of a graph embedding vector generation method according to the present disclosure. The graph embedding vector generation method may include the following steps:

[0089] Step 401: Obtain the target user's behavioral information regarding the information, whereby the behavioral information includes the target query information and the behavioral type information.

[0090] In this embodiment, the execution body of the graph embedding vector generation method (e.g.) Figure 1 The terminal devices 101, 102, and 103 shown can obtain user behavior information related to information locally, or the execution entity of the graph embedding vector generation method (e.g., Figure 1The server 105 shown is located locally or via a network (e.g., Figure 1 The network 104 shown above starts from the terminal device (e.g., Figure 1 The terminal devices 101, 102, and 103 shown acquire user behavior information related to the information. This behavior information can be the information corresponding to the actions taken by the target user on the retrieved information.

[0091] Step 402: Construct an information relevance graph based on the behavioral information. The nodes of the information relevance graph include target user nodes and target query information nodes. The weights of the edges of the information relevance graph are determined based on the behavioral type information.

[0092] In this embodiment, the aforementioned executing entity can construct an information correlation graph based on behavioral information. This information correlation graph can include nodes and edges. Nodes can include target user nodes and target query information nodes, and the weights of the edges can be determined based on behavioral type information. The behavioral type information can be information corresponding to the type of operation performed by the target user on the information, such as the timing of the operation, the frequency of the operation, and the type of the operation. The type of operation can include at least one of the following: clicking, browsing, forwarding, etc.

[0093] Step 403: Perform graph embedding on the information correlation graph to obtain the feature vectors corresponding to the target user node and the target query information node in the information correlation graph.

[0094] In this embodiment, the aforementioned execution entity can perform graph embedding on the information-related graph to obtain the feature vectors corresponding to the target user node and the target query information node in the information-related graph. The above-mentioned graph embedding is also known as network embedding or network representation learning. Its purpose is to project each node in the information-related graph into a low-dimensional vector space, that is, to learn an effective representation or encoding for the information-related graph. These representations or encodings are the "embedding" of the information-related graph.

[0095] Step 404: Obtain the first query information of the first user.

[0096] In this embodiment, the aforementioned execution entity (e.g. Figure 1 The terminal devices 101, 102, and 103 shown can obtain the first query information of the first user; or the aforementioned execution entity (e.g. Figure 1 The server 105 shown communicates via a network (e.g., Figure 1 The network 104 shown obtains first query information input by a first user on a terminal device. This first query information can be query information input by the first user for use in pushing information. The first user can be a target user or a new user other than the target user.

[0097] Step 405: Determine the matching degree between the target query information and the first user's first query information.

[0098] In this embodiment, the execution entity can calculate the matching degree between the target query information and the first user's first query information. This matching degree can be the similarity between the target query information and the first query information, for example, the similarity between entity words in the target query information and entity words in the first query information.

[0099] Step 406: In response to the matching degree meeting the preset matching degree threshold, the target information corresponding to the target user is pushed to the first user.

[0100] In this embodiment, the execution entity can push target information corresponding to the target user to the first user when the matching degree meets a preset matching degree threshold. "Meeting" here refers to the matching degree meeting the preset matching degree threshold.

[0101] It should be noted that the above-mentioned preset matching threshold can be set based on the accuracy of information push or by the user based on work experience, which will not be elaborated here.

[0102] In this embodiment, the specific operations of steps 401, 402, and 403 have been described. Figure 2 Steps 201, 202, and 203 in the illustrated embodiments are described in detail and will not be repeated here.

[0103] from Figure 4 It can be seen from this that, with Figure 2 Compared to the corresponding embodiments, the graph embedding vector generation method in this embodiment emphasizes the step of pushing information to the first user based on the matching degree between the first user's first query information and the target user's target query information. Therefore, the scheme described in this embodiment can obtain the first user's first query information; then determine the matching degree between the target query information and the first user's first query information; and finally, in response to the matching degree meeting a preset matching degree threshold, push the target information corresponding to the target user to the first user. This process, which can determine the most matching information for the first user based on the information relevance graph, thereby promoting user click-through and conversion behaviors, effectively improves the information recall rate, resulting in a significant improvement in the ROI of information delivery.

[0104] Further reference Figure 5 , Figure 5 This illustrates an application scenario where targeted information is delivered to a first user. In this application scenario, server 502 (e.g., Figure 1 The server 105 shown can be accessed from a terminal device (e.g., Figure 1The terminal devices 101, 102, and 103 shown acquire behavioral information 503 of target users (e.g., multiple target users) regarding information, whereby the behavioral information includes target query information and behavioral type information. Then, server 502 constructs an information correlation graph 504 based on the behavioral information 503. Next, graph embedding 505 is performed on the information correlation graph to obtain feature vectors corresponding to target user nodes and target query information nodes in the information correlation graph. Finally, the matching degree between the target query information and the acquired first query information of the first user is determined. When the matching degree meets a preset matching threshold, the target information corresponding to the target user is pushed to the first user 507.

[0105] In this application scenario, the ability to determine the most relevant information between the first user and the target user based on the information correlation graph can effectively improve the recall rate of information, thereby significantly improving the ROI of information delivery.

[0106] Further reference Figure 6 , Figure 6 An application scenario of the graph embedding vector generation method according to this disclosure is illustrated. In this application scenario, the graph embedding vector generation method may include the following steps:

[0107] The first step is to obtain the target user's behavioral information regarding the information, which includes the target query information and the behavior type information.

[0108] The second step is to categorize the target query information according to the business type to which it belongs.

[0109] The third step is to extract the core keywords from the categorized target query information.

[0110] The fourth step is to determine the weights of the edges in the information-related graph.

[0111] The fifth step is to construct an information correlation graph based on the core keywords and behavior type information in the categorized target query information.

[0112] The sixth step is to perform graph embedding on the information-related graph to obtain the feature vectors corresponding to the target user node and the core word (i.e., the core words in the classified target query information) nodes.

[0113] Step 7: Obtain the first query information of the first user and determine whether the first user is the target user.

[0114] Step 8: Determine the matching degree between the core words in the target query information and the core words in the first query information; when the matching degree meets the preset matching degree threshold, push the target information corresponding to the target user to the first user.

[0115] Further reference Figure 7As an implementation of the methods shown in the above figures, this disclosure provides an embodiment of a graph embedding vector generation apparatus, which is similar to... Figure 2 Corresponding to the method embodiments shown, this device can be specifically applied to various electronic devices.

[0116] like Figure 7 As shown, the graph embedding vector generation device 700 of this embodiment may include: an information acquisition module 701, a graph construction module 702, and a graph embedding module 703. The information acquisition module 701 is configured to acquire behavioral information of the target user regarding information, wherein the behavioral information includes target query information and behavioral type information; the graph construction module 702 is configured to construct an information-related graph based on the behavioral information, wherein the nodes of the information-related graph include target user nodes and target query information nodes, and the weights of the edges of the information-related graph are determined based on the behavioral type information; the graph embedding module 703 is configured to perform graph embedding on the information-related graph to obtain feature vectors corresponding to the target user nodes and target query information nodes in the information-related graph.

[0117] In this embodiment, the specific processing of the information acquisition module 701, the graph construction module 702, and the graph embedding module 703 in the graph embedding vector generation device 700, and the resulting technical effects, can be found in reference to [reference needed]. Figure 2 The relevant descriptions of steps 201-203 in the corresponding embodiments will not be repeated here.

[0118] In some optional implementations of this embodiment, the graph embedding module 703 includes: a first obtaining unit configured to perform a random walk on the information correlation graph to obtain a sequence of behaviors of the target user toward the target query information; and a second obtaining unit configured to perform vector embedding on the behavior sequence to obtain feature vectors corresponding to the target user node and the target query information node in the information correlation graph.

[0119] In some optional implementations of this embodiment, the first obtaining unit is further configured to: divide the information correlation graph into multiple target information correlation graphs; and for each target information correlation graph, perform a random walk on each target information correlation graph to obtain the target user's behavior sequence for the target query information.

[0120] In some optional implementations of this embodiment, the behavior type information includes at least one of the following: the timing of the operation, the frequency of the operation, and the type of the operation.

[0121] In some optional implementations of this embodiment, if the behavior type information includes the timing of the operation, the frequency of the operation, and the type of the operation; and the graph embedding vector generation device further includes: a weight determination module, configured to determine the weights of the edges of the information-related graph based on the weights corresponding to the timing of the operation, the frequency of the operation, and the type of the operation.

[0122] In some optional implementations of this embodiment, the graph embedding vector generation device further includes: an information acquisition module 701, configured to acquire first query information of a first user; a matching degree determination module, configured to determine the matching degree between the target query information and the first query information of the first user; and an information push module, configured to push the target information corresponding to the target user to the first user in response to the matching degree meeting a preset matching degree threshold.

[0123] In some optional implementations of this embodiment, before constructing the information correlation graph, the graph embedding vector generation device further includes: an information classification module, configured to classify the target query information according to the business type to which the target query information belongs; and a graph construction module 702, further configured to construct an information correlation graph based on the classified query information and behavior type information.

[0124] In some optional implementations of this embodiment, the information classification module includes: a business type determination unit, configured to determine the business type to which the target query information belongs based on a pre-trained classification model; and an information classification unit, configured to classify the target query information according to the business type to which the target query information belongs.

[0125] In some optional implementations of this embodiment, the graph embedding vector generation device further includes: an information acquisition module 701, configured to acquire entity requirement limitation information in response to the character length of the target query information meeting a preset length threshold; and a model optimization module, configured to use the entity requirement limitation information and the target query information as training samples to optimize and train a pre-trained classification model.

[0126] In some optional implementations of this embodiment, the target query information includes core keywords.

[0127] According to embodiments of this disclosure, this disclosure also provides an electronic device, a readable storage medium, and a computer program product.

[0128] Figure 8A schematic block diagram of an example electronic device 800 that can be used to implement embodiments of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0129] like Figure 8 As shown, device 800 includes a computing unit 801, which can perform various appropriate actions and processes based on a computer program stored in read-only memory (ROM) 802 or a computer program loaded from storage unit 808 into random access memory (RAM) 803. RAM 803 may also store various programs and data required for the operation of device 800. The computing unit 801, ROM 802, and RAM 803 are interconnected via bus 804. Input / output (I / O) interface 805 is also connected to bus 804.

[0130] Multiple components in device 800 are connected to I / O interface 805, including: input unit 806, such as keyboard, mouse, etc.; output unit 807, such as various types of monitors, speakers, etc.; storage unit 808, such as disk, optical disk, etc.; and communication unit 809, such as network card, modem, wireless transceiver, etc. Communication unit 809 allows device 800 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0131] The computing unit 801 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 801 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 801 performs the various methods and processes described above, such as the graph embedding vector generation method. For example, in some embodiments, the graph embedding vector generation method can be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 808. In some embodiments, part or all of the computer program can be loaded and / or installed on device 800 via ROM 802 and / or communication unit 809. When the computer program is loaded into RAM 803 and executed by the computing unit 801, one or more steps of the graph embedding vector generation method described above can be performed. Alternatively, in other embodiments, the computing unit 801 can be configured to perform the graph embedding vector generation method by any other suitable means (e.g., by means of firmware).

[0132] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0133] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0134] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. Machine-readable media can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0135] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0136] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with embodiments of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0137] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact via communication networks. Client-server relationships are created by computer programs running on the respective computers and having a client-server relationship with each other. Servers can be cloud servers, servers in distributed systems, or servers incorporating blockchain technology.

[0138] Artificial intelligence (AI) is the study of how computers can simulate certain human thought processes and intelligent behaviors (such as learning, reasoning, thinking, and planning). It encompasses both hardware and software technologies. AI hardware technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, and big data processing. AI software technologies mainly include computer vision, speech recognition, natural language processing, machine learning / deep learning, big data processing, and knowledge graph technologies.

[0139] It should be understood that the various forms of processes shown above can be used to reorder, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution mentioned in this disclosure can be achieved, and this is not limited herein.

[0140] The specific embodiments described above do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A method for generating graph embedding vectors, comprising: Obtain target user behavior information regarding information, wherein the behavior information includes target query information and behavior type information, the target query information is the information input by the target user to query information, and the behavior type information is the information corresponding to the operation performed on the queried information; The target query information is categorized according to the business type to which it belongs; Based on the classified target query information and the behavior type information, an information relevance graph is constructed, wherein the nodes of the information relevance graph include target user nodes and target query information nodes, and the weights of the edges of the information relevance graph are determined based on the behavior type information; A random walk is performed on the information correlation graph to obtain the target user's behavior sequence for the target query information. The behavior sequence is then vector-embedded to obtain the feature vectors corresponding to the target user node and the target query information node in the information correlation graph.

2. The method according to claim 1, wherein, The step of performing a random walk on the information correlation graph to obtain the target user's behavioral sequence for the target query information includes: The information correlation graph is divided into multiple target information correlation graphs; For each target information correlation graph, a random walk is performed on each target information correlation graph to obtain the target user's behavior sequence for target query information.

3. The method according to claim 2, wherein, The behavior type information includes at least one of the following: The timing of operations, the frequency of operations, and the type of operations.

4. The method according to claim 3, wherein, If the behavior type information includes the timing of the operation, the frequency of the operation, and the type of the operation; the weights of the edges in the information-related graph are determined based on the following steps: The weights of the edges in the information-related graph are determined based on the weights corresponding to the timing, frequency, and type of the operations.

5. The method according to claim 4, further comprising: Retrieve the first user's first query information; Determine the matching degree between the target query information and the first query information of the first user; In response to the matching degree meeting a preset matching degree threshold, the target information corresponding to the target user is pushed to the first user.

6. The method according to claim 1, wherein, The step of classifying the target query information according to the business type to which the target query information belongs includes: Based on a pre-trained classification model, determine the business type to which the target query information belongs; The target query information is categorized according to the business type to which it belongs.

7. The method according to claim 6, wherein, The pre-trained classification model is determined based on the following steps: In response to the target query information having a character length that meets a preset length threshold, entity requirement limitation information is obtained; The entity demand constraint information and target query information are used as training samples to optimize the pre-trained classification model.

8. The method according to any one of claims 1-7, wherein, The target query information includes core keywords.

9. A graph embedding vector generation apparatus, comprising: The information acquisition module is configured to acquire the behavioral information of the target user regarding the information. The behavioral information includes target query information and behavior type information. The target query information is the information input by the target user to query the information, and the behavior type information is the information corresponding to the operation performed on the queried information. The information classification module is configured to classify the target query information according to the business type to which the target query information belongs; The graph construction module is configured to construct an information relevance graph based on the classified target query information and the behavior type information, wherein the nodes of the information relevance graph include target user nodes and target query information nodes, and the weights of the edges of the information relevance graph are determined based on the behavior type information; The graph embedding module is configured to perform graph embedding on the information-related graph to obtain feature vectors corresponding to target user nodes and target query information nodes in the information-related graph. The module includes: a first obtaining unit configured to perform a random walk on the information-related graph to obtain a sequence of target user behavior towards target query information; and a second obtaining unit configured to perform vector embedding on the behavior sequence to obtain feature vectors corresponding to target user nodes and target query information nodes in the information-related graph.

10. The apparatus according to claim 9, wherein, The first obtaining unit is further configured to: The information correlation graph is divided into multiple target information correlation graphs; For each target information correlation graph, a random walk is performed on each target information correlation graph to obtain the target user's behavior sequence for target query information.

11. The apparatus according to claim 10, wherein, The behavior type information includes at least one of the following: The timing of operations, the frequency of operations, and the type of operations.

12. The apparatus according to claim 11, wherein, If the behavior type information includes the timing of the operation, the frequency of the operation, and the type of the operation; the device further includes: The weight determination module is configured to determine the weights of the edges of the information-related graph based on the weights corresponding to the timing, frequency, and type of the operation.

13. The apparatus of claim 12, further comprising: The information acquisition module is configured to acquire the first query information of the first user; The matching degree determination module is configured to determine the matching degree between the target query information and the first query information of the first user; The information push module is configured to push target information corresponding to the target user to the first user in response to the matching degree meeting a preset matching degree threshold.

14. The apparatus according to claim 9, wherein, The information classification module includes: The business type determination unit is configured to determine the business type to which the target query information belongs based on a pre-trained classification model; The information classification unit is configured to classify the target query information according to the business type to which the target query information belongs.

15. The apparatus of claim 14, further comprising: The information acquisition module is configured to acquire entity requirement limitation information in response to the target query information having a character length that meets a preset length threshold. The model optimization module is configured to use entity requirement constraint information and target query information as training samples to optimize the pre-trained classification model.

16. The apparatus according to any one of claims 9-15, wherein, The target query information includes core keywords.

17. An electronic device comprising: At least one processor; as well as A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-8.

18. A non-transitory computer-readable storage medium storing computer instructions for causing the computer to perform the method of any one of claims 1-8.

19. A computer program product comprising a computer program that, when executed by a processor, implements the method according to any one of claims 1-8.

Images